AI Models for the Study of Motion Perception

Article: Artificial ‘brain’ reveals why we can’t always believe our eyes
Source: University of Cambridge, via Technology Networks
Published: February 25, 2021

Researchers at the University of Cambridge have developed an artificial neural network to study motion perception. Called MotionNet, this computer network simulates a particular type of visual processing known as the reverse phi phenomenon. In the phi phenomenon, which most of us are familiar with in the form of animated films, dark spots appearing in succession give the illusion of motion. In the more perplexing reverse phi phenomenon, if the second point becomes light rather than dark, then we perceive the motion as moving in the opposite direction; instead of perceiving the sequence moving "forward," we perceive it as moving "backward" (keeping in mind that directions are relative). The MotionNet system seems to faithfully replicate the mistakes that human brains make with regard to the reverse phi illusion, but has the advantage of being able to be examined and tested in detail. The researchers found, for example, that the reverse phi illusion triggered "neurons" in their system that were tuned to the direction opposite of the actual movement. The system also revealed information about the speed of movement and spacing of dots on the effect of motion perception. Studying optical illusions has consequential implications for patient care beyond scholarly interest. For example, in previous work, the researchers showed that neurons in our brain are biased towards slow speeds, so when visibility is low we tend to think that objects are moving more slowly than they actually are. These findings are thus applicable to real-world scenarios, such as accurately gauging the speed of moving objects when driving in low visibility conditions. The MotionNet artificial neural network is a model wherein preliminary experiments could provide insights for more focused studies in subsequent biological models and human subjects.

My rating of this study: ⭐⭐

See also: Reverse phi motion in fruit flies